Add docs about redundancy (#2142)

## Issue Addressed

- Resolves #2140

## Proposed Changes

Adds some documentation on the topic of "redundancy".

## Additional Info

NA
This commit is contained in:
Paul Hauner 2021-01-12 00:26:22 +00:00
parent 423dea169c
commit 1d535659d6
3 changed files with 123 additions and 0 deletions

View File

@ -33,6 +33,7 @@
* [Local Testnets](./local-testnets.md)
* [Advanced Networking](./advanced_networking.md)
* [Running a Slasher](./slasher.md)
* [Redundancy](./redundancy.md)
* [Contributing](./contributing.md)
* [Development Environment](./setup.md)
* [FAQs](./faq.md)

View File

@ -174,3 +174,11 @@ or after being off for more than several minutes.
If this log continues appearing sporadically during operation, there may be an
issue with your eth1 endpoint.
### Can I use redundancy in my staking setup?
You should **never** use duplicate/redundant validator keypairs or validator clients (i.e., don't
duplicate your JSON keystores and don't run `lighthouse vc` twice). This will lead to slashing.
However, there are some components which can be configured with redundancy. See the
[Redundancy](./redundancy.md) guide for more information.

114
book/src/redundancy.md Normal file
View File

@ -0,0 +1,114 @@
# Redundancy
[subscribe-api]: https://ethereum.github.io/eth2.0-APIs/#/Validator/prepareBeaconCommitteeSubnet
There are three places in Lighthouse where redundancy is notable:
1. ✅ GOOD: Using a redundant Beacon node in `lighthouse bn --beacon-nodes`
1. ✅ GOOD: Using a redundant Eth1 node in `lighthouse bn --eth1-endpoints`
1. ☠️ BAD: Running redundant `lighthouse vc` instances with overlapping keypairs.
I mention (3) since it is unsafe and should not be confused with the other two
uses of redundancy. **Running the same validator keypair in more than one
validator client (Lighthouse, or otherwise) will eventually lead to slashing.**
See [Slashing Protection](./slashing-protection.md) for more information.
From this paragraph, this document will *only* refer to the first two items (1, 2). We
*never* recommend that users implement redundancy for validator keypairs.
## Redundant Beacon Nodes
The `lighthouse bn --beacon-nodes` flag allows one or more comma-separated values:
1. `lighthouse vc --beacon-nodes http://localhost:5052`
1. `lighthouse vc --beacon-nodes http://localhost:5052,http://192.168.1.1:5052`
In the first example, the validator client will attempt to contact
`http://localhost:5052` to perform duties. If that node is not contactable, not
synced or unable to serve the request then the validator client may fail to
perform some duty (e.g., produce a block or attest).
However, in the second example, any failure on `http://localhost:5052` will be
followed by a second attempt using `http://192.168.1.1:5052`. This
achieves *redundancy*, allowing the validator client to continue to perform its
duties as long as *at least one* of the beacon nodes is available.
There are a few interesting properties about the list of `--beacon-nodes`:
- *Ordering matters*: the validator client prefers a beacon node that is
earlier in the list.
- *Synced is preferred*: the validator client prefers a synced beacon node over
one that is still syncing.
- *Failure is sticky*: if a beacon node fails, it will be flagged as offline
and wont be retried again for the rest of the slot (12 seconds). This helps prevent the impact
of time-outs and other lengthy errors.
> Note: When supplying multiple beacon nodes the `http://localhost:5052` address must be explicitly
> provided (if it is desired). It will only be used as default if no `--beacon-nodes` flag is
> provided at all.
### Configuring a redundant Beacon Node
In our previous example we listed `http://192.168.1.1:5052` as a redundant
node. Apart from having sufficient resources, the backup node should have the
following flags:
- `--staking`: starts the HTTP API server and ensures the Eth1 chain is synced.
- `--http-address 0.0.0.0`: this allows *any* external IP address to access the
HTTP server (a firewall should be configured to deny unauthorized access to port
`5052`). This is only required if your backup node is on a different host.
- `--subscribe-all-subnets`: ensures that the beacon node subscribes to *all*
subnets, not just on-demand requests from validators.
- `--process-all-attestations`: ensures that the beacon node performs
aggregation on all seen attestations.
Subsequently, one could use the following command to provide a backup beacon
node:
```bash
lighthouse bn \
--staking \
--http-address 0.0.0.0 \
--subscribe-all-subnets \
--process-all-attestations
```
### Resource usage of redundant Beacon Nodes
The `--subscribe-all-subnets` and `--process-all-attestations` flags typically
cause a significant increase in resource consumption. A doubling in CPU
utilization and RAM consumption is expected.
The increase in resource consumption is due to the fact that the beacon node is
now processing, validating, aggregating and forwarding *all* attestations,
whereas previously it was likely only doing a fraction of this work. Without
these flags, subscription to attestation subnets and aggregation of
attestations is only performed for validators which [explicitly request
subscriptions](subscribe-api).
There are 64 subnets and each validator will result in a subscription to *at
least* one subnet. So, using the two aforementioned flags will result in
resource consumption akin to running 64+ validators.
## Redundant Eth1 nodes
Compared to redundancy in beacon nodes (see above), using redundant Eth1 nodes
is very straight-forward:
1. `lighthouse bn --eth1-endpoints http://localhost:8545`
1. `lighthouse bn --eth1-endpoints http://localhost:8545,http://192.168.0.1:8545`
In the case of (1), any failure on `http://localhost:8545` will result in a
failure to update the Eth1 cache in the beacon node. Consistent failure over a
period of hours may result in a failure in block production.
However, in the case of (2), the `http://192.168.0.1:8545` Eth1 endpoint will
be tried each time the first fails. Eth1 endpoints will be tried from first to
last in the list, until a successful response is obtained.
There is no need for special configuration on the Eth1 endpoint, all endpoints can (probably should)
be configured identically.
> Note: When supplying multiple endpoints the `http://localhost:8545` address must be explicitly
> provided (if it is desired). It will only be used as default if no `--eth1-endpoints` flag is
> provided at all.